(Nanowerk News) The development of neural networks to create artificial intelligence in computers was originally inspired by how biological systems work. This neuromorphic network, however, runs on hardware that doesn’t look like a biological brain, which limits performance. Now, researchers from Osaka University and Hokkaido University plan to change that by creating neuromorphic wetware.
The research has been published in Advanced Functional Materials (“Creation and Training of 3D Conductive Polymer Networks for Neuromorphic Wetware”).
While neural network models have achieved tremendous success in applications such as image generation and cancer diagnosis, they still lag far behind the general processing capabilities of the human brain. This is partly because it is implemented in software using traditional computer hardware that is not optimized for the millions of parameters and connections that these models typically require.
Neuromorphic wetware, based on memristive devices, can overcome this problem. A memristive device is a device whose resistance is determined by the history of the applied voltage and current. In this approach, electropolymerization is used to connect electrodes immersed in the precursor solution using wires made of conductive polymers. The resistance of each wire is then adjusted using tiny voltage pulses, producing a memristive device.
“The potential for creating fast and energy-efficient networks has been demonstrated using 1D or 2D structures,” said senior author Megumi Akai-Kasaya. “Our goal is to extend this approach to building 3D networks.”
The researchers were able to grow polymer cables from a common polymer blend called ‘PEDOT:PSS’, which is highly conductive, transparent, flexible and stable. The 3D structures of the top and bottom electrodes were first immersed in the precursor solution. PEDOT:PSS wires are then implanted between selected electrodes by applying a square wave voltage to these electrodes, mimicking the formation of synaptic connections via axon guidance in an immature brain.
Once the wire is formed, the characteristics of the wire, particularly the conductance, are controlled using tiny voltage pulses applied to a single electrode, which change the electrical properties of the film surrounding the wire.
“The process is continuous and reversible,” explains lead author Naruki Hagiwara, “and it is these characteristics that allow networks to be trained, such as software-based neural networks.”
Artificial networks were used to demonstrate unsupervised Hebbian learning (that is, when synapses that frequently fire together strengthen their shared connections over time). What’s more, the researchers can precisely control the conductance value of the cable so that the network can complete its task. Spike-based learning, another approach to neural networks that more closely resembles biological neural network processes, was also demonstrated by controlling the diameter and conductivity of the wires.
Next, by fabricating chips with a greater number of electrodes and using microfluidic channels to supply precursor solutions to each electrode, the researchers hope to build larger and more robust networks. Overall, the approach defined in this study represents a major step towards the realization of neuromorphic wet devices and closing the gap between human and computer cognitive abilities.